I have a P43G (PCChips) motherboard, with a E3300 Celeron 775, running with 4gb of DDR800 memory, dual channel, running at 6-6-6-18. The processor is running at the stock 2.5 gHz.

After running for a day or so, the screen goes nuts, not updating all portions. I run Slackware 13.37 on this system, with KDE. If I restart X windows things work OK for another day or so. A few times, I had to reboot, and /var/log/messages showed that there was a GPU failure, but it was not evident what type or cause of the failure.

I use this machine daily, so I am running Memtest86 overnight, but as I expected there have been no memory errors as of yet. I also opened up the box, because in the summer, under humid conditions, I can get similar flakiness with dust bunnies in the box. Everything was vac'd out, but it wasn't that dirty.

Suggestions are welcome as to how I can find the cause, and any guidance as to what I should be chasing. Thanks.

You didnt mention which video chip/card do you have, that might be important info, also what driver is installed for the video?
If this is your mobo then this: Integrated Intel Graphics Media Accelerator 3100 (GMA3100) Is your video if you are using the integrated one, please confirm.
Also post the output for:
Regards

82g33/g31

When the screen goes nuts, check the temperatures and the voltages, particularly 5V & 3.3V. They're often in the bios under system health check

The temps and voltages were OK, running BIOS, but that does not address the dynamic state.

Running Firefox, I have some partially corrupted icons, so the problem may have another origin. I will find good motherboard test points for the voltages, and probe in there when I have having more corruption.

The temps are good, as the box is very well cooled (4 fans, cool room, etc.) and PIR temp readings are low. Sometimes the processor does get a bit on the high side, however. But the graphics are cool. Last time I did the graphics chips, I got an area reading of 59C.

If you are using the S3 X-driver, then that is familiar.
I have similar problem with that driver not updating all parts of screen. The problem
will only show up after using the machine awhile. It is obvious on certain parts of the window frames and their controls.
It is affected by recompiling the kernel with different options, but I cannot figure out why.
Thought the huge kernel was not affected but am not sure.
It is apparent that some part of the graphics memory cache eventually gets clobbered.
No problem with same hardware running XP, so I suspect that intel driver does not handle
graphics memory correctly for something that X is doing.
Also suspect that their may be a kernel module incompatibility.

Consider module marking option so only modules compiled for that particular kernel can load.
That fixed a problem with video drivers on another machine here.

Most video drivers have to be compiled on the modified kernel that they will run with.
Some kernel data structures can be different size or missing when kernel options are different than in the kernel that compiled the video module.
Using driver modules on mostly similar kernels will have effects like this.
It looks good for a while, sometimes for hours, until misaligned usage of kernel data structures finally blows up the driver execution.

Your Memtest probably does not test graphics memory.

Thanks for the suggestion and experience with the S3 driver, and I will check to see if I am using that (not sure right now). I kinda doubt it, as the graphics is G31, but I will investigate.

I am using the huge kernel.

memtest86+ does not text graphics memory, but then again, this MB and processor configuration uses shared memory, and memtest does check that, except for a 256k buffer area used for the display when running memtest.

Again, I am using the onboard motherboard G31 graphics.

Thanks for your ideas.

I was thinking Intel, but this machine using S3 driver has SavagePro, but the problems are
similar.
After posting, I finally changed it to the VESA X-driver, and the problem went away.

I have done game graphics programming and it looks like the textures kept in graphics memory
were being overwritten by something else. If this is the same appearance that you are getting, then you should suspect software, not hardware.

This reminds me that X-drivers do not obey rules setup by the console drivers or Linux drivers. It could be another driver using the graphics memory at the same time that
the X-driver is storing textures in it.

On another machine, KDE would crash (but XFCE4 would work) when using the nvidia X-driver, and recently changing to using the Nvidia commercial driver instead has fixed many problems.

My box threw me a line recently

"Graphics programming in X is like calculating the exact value of pi using Roman numerals" (Somebody's quote, not reproduced accurately)

Is that right?